Обсуждение: Comparing two double values method

Dears,

I noticed that in the `check_GUC_init` function, there is a direct
comparison using the != operator for two double values, which seems
problematic.

I wrote this patch to fix this.

--
Regard,
Bowen Shi

On Tue, 10 Oct 2023 at 12:33, Bowen Shi <zxwsbg12138@gmail.com> wrote:
>
> Dears,
>
> I noticed that in the `check_GUC_init` function, there is a direct
> comparison using the != operator for two double values, which seems
> problematic.

I don't think I understand the problem. The code checks that the
dynamic initialization values are equal to the current value of the
GUC, or 0. Why would a "margin for error" of 1e-6 be of any use?
Why was the margin of 1e-6 chosen instead of one based on the exponent
of the GUC's current value (if any)?

In my view, this would break the code, not fix it, as it would
decrease the cases where we detect broken GUC registrations.

Kind regards,

Matthias van de Meent
Neon (https://neon.tech)

On 10/10/2023 13:31, Bowen Shi wrote:
> Dears,
> 
> I noticed that in the `check_GUC_init` function, there is a direct
> comparison using the != operator for two double values, which seems
> problematic.
> 
> I wrote this patch to fix this.

No, the compile-time initial values should match exactly.

-- 
Heikki Linnakangas
Neon (https://neon.tech)

You're right, I made a mistake.

Thanks for your explanation.

Heikki Linnakangas <hlinnaka@iki.fi> writes:
> On 10/10/2023 13:31, Bowen Shi wrote:
>> I noticed that in the `check_GUC_init` function, there is a direct
>> comparison using the != operator for two double values, which seems
>> problematic.

> No, the compile-time initial values should match exactly.

Right.  The point of this test is to catch cases where you wrote,
say,

    double my_guc = 1.1;

but the boot_val for it in guc_tables.c is 1.2.  There is no
reason to allow any divergence in the spellings of the two C
literals, so as long as they're compiled by the same compiler
there's no reason to expect that the compiled values wouldn't
be bit-equal.

The point of the exclusions for zero is to allow you to just
write

    double my_guc;

without expressing an opinion about the initial value.
(Yes, this does mean that "double my_guc = 0.0;" could
be misleading.  It's just a heuristic though.)

            regards, tom lane